Share with your CEO
Boards are asking “what’s our AI strategy?” when they should be asking “what’s our business strategy, and where does AI serve it?” That reframe, pressed by Julie Averill, former CIO of lululemon turned board director, and Dan French of Consider Solutions, cuts through what Averill calls AI “whitewashing,” boards performing fluency in a technology they don’t actually govern. The core diagnostic from French: only 30 to 40% of business problems genuinely benefit from AI’s probabilistic approach; deploying it everywhere is a governance failure as much as a technical one.
What this means for your business
The companies most exposed here aren’t the laggards who haven’t deployed AI yet. They’re the ones that have deployed it enthusiastically, measured success by adoption rates, and handed governance to a committee rather than to the functional leaders actually running the decisions. If your board can answer “how many employees are using AI tools” faster than it can answer “where is AI making consequential decisions in this company,” you have the whitewashing problem Averill is describing.
French’s 30 to 40% figure deserves to sit in more board decks. AI, at its technical core, works on probabilities, generating the most statistically likely output rather than executing a fixed rule. That’s genuinely powerful for demand forecasting, candidate screening, or customer churn modeling, cases where pattern recognition beats rigid logic. It’s a liability for payroll calculations, regulatory filings, or contract execution, where you need the same answer every time. Most organizations haven’t done that triage deliberately. They’ve let tool availability drive deployment rather than letting problem type drive tool selection, and that inversion is where audit risk quietly accumulates.
Averill’s accountability model is the piece most boards will resist but most need. Her argument is that AI governance belongs inside each function, with the CFO owning every agent touching the finance stack and the supply chain leader owning any decision an algorithm makes about inventory routing. That’s harder to operationalize than a central AI governance council, but it’s also the only model that survives at scale. Central councils become bottlenecks; diffused accountability becomes nobody’s problem. The falsification condition is simple: if your company can’t name a human owner for each AI-driven decision today, the council model is already failing you.
Concept deep-dive: Deterministic vs. probabilistic systems
Deterministic software follows fixed rules and returns the same output for the same input every time, think a payroll engine that always calculates overtime the same way. Probabilistic systems, which is what AI models are, generate outputs based on statistical likelihood, the best guess given patterns in training data. That distinction matters for governance because probabilistic outputs require human judgment to catch edge cases, whereas deterministic outputs can be audited by checking whether the rule was applied correctly.
Based on reporting from The AI governance question boards are getting wrong, originally published 2026-07-16 10:56:00.
